About a century ago, a solar eclipse quietly overthrew Isaac Newton. Astronomers pointed telescopes at the sun, checked the starlight, and found a tiny mismatch with his prediction—just enough to crown Einstein instead. This episode asks: what lets an idea lose that clearly?
Feynman said science is a way of trying *not* to fool yourself—but in daily life, most of us run the opposite experiment. We collect confirming evidence, avoid hard tests, and quietly move the goalposts when reality disagrees. That’s how diets become “it works if you do it right,” business strategies become “the timing just wasn’t right,” and pet theories about people turn into “you don’t really know them like I do.”
This episode is about a simple but ruthless question: “What observation would make me admit I’m wrong?” Popper argued that until you can answer that, you’re not really doing science—you’re just telling stories. We’ll see how falsifiability shapes real research, why preregistered studies changed parts of psychology, how drug trials are built to break promising ideas, and how you can borrow the same mindset to stress‑test your own beliefs before the world does it for you.
In real life, though, we rarely ask that sharp “what would prove me wrong?” question up front. We start a new habit because it “feels right,” back a business idea because a friend succeeded with something similar, or trust a personality test because the description sounds uncannily accurate. Then, once we’re invested, we quietly reinterpret every outcome as support. A bad week means the habit “hasn’t kicked in yet,” slow sales mean “the market isn’t educated,” a wrong prediction becomes “basically right, just off on timing.” Without clear failure points, every result turns into a win—and we stop learning.
Popper’s sharp move was to stop asking “Is this idea reasonable?” and start asking “Where does this idea stick its neck out?” An idea can be wild and still be scientific, as long as it risks a clear collision with reality. That’s why “this treatment reduces blood pressure by 10 points on average within 8 weeks” belongs in a lab, while “everything happens for a reason” doesn’t. One tells you exactly what numbers to look for; the other can flex to fit anything that happens.
This is also why conspiracy theories are so hard to kill. They often come wrapped in escape hatches: any evidence against them “proves” how powerful the hidden forces are. Ask, “What evidence would convince you this isn’t true?” and the answer is usually: “There isn’t any.” That’s a red flag—not that the claim is false, but that it has no built‑in way to lose.
The 2018 replication shock in psychology exposed a quieter version of the same problem. When hypotheses are mushy—“people are influenced by subtle cues”—almost any pattern in noisy data can be spun as a win. Researchers weren’t all cheating; they were working with claims that could bend instead of break. Turn that into: “Participants shown primes A, B, C will score at least 0.3 standard deviations higher on measure X than those shown neutral primes,” and suddenly lots of those “wins” risk becoming clean losses.
Good theories don’t just survive tests; they generate *risky* tests. General relativity didn’t win because it was elegant—it won because it dared to say, “Measure this angle, at this time, and you’ll see this much difference.” When the numbers lined up and Newton’s didn’t, physicists didn’t say “close enough”; they updated.
In everyday life, we tend to do the opposite: we fit our stories to the outcome. A manager launches a new process, performance doesn’t improve, and the story becomes “well, it prevented things from being even worse.” A relationship pattern hurts, but we tell ourselves it “shows how deeply we care.” Without prior, specific stakes—“if X hasn’t changed by Y date, I’ll call this wrong”—we rewrite the script rather than question the plot.
The practical habit is to front‑load the discomfort: tie each belief to a small, concrete bet with the world, before the results come in.
Think about claims you hear all the time: “This meeting format makes us more productive,” “Cutting carbs boosts my focus,” “Our new onboarding keeps people longer.” They sound plausible, but they rarely come with a clear way to lose. What would disconfirm each one?
You could say: “If average task completion time hasn’t dropped 15% after four weeks of this meeting style, we’ll scrap it.” Or: “If my afternoon concentration isn’t noticeably better on at least 10 of 14 days (rated blind, using a simple 1–5 scale I fill out before checking what I ate), I’ll stop insisting carbs are the culprit.” For onboarding: “If 6‑month retention doesn’t improve by at least 5 percentage points over last year’s cohort, we’ll admit this version didn’t help.”
A good scientific hypothesis is like a cooking recipe that specifies what the dish should taste like; once you serve it, the group either says, “yes, this is it,” or “no, this missed the mark,” and you change the recipe instead of blaming the guests.
“Science is a way of trying not to fool yourself” applies far beyond labs. Future workplaces may treat strategies like software versions: launch 1.0 with explicit “kill metrics,” then auto‑retire or revise when results fall short. In classrooms, students could tag claims in essays with “here’s what would prove me wrong,” normalizing revision instead of defensiveness. Online, platforms might flag posts lacking any testable core, nudging us toward claims that can, in principle, be checked.
Treat “what would prove me wrong?” as a quiet compass, not a weapon. Aim it at career plans, political takes, even your picture of yourself. Let friends act as honest mirrors, like coaches timing your sprints instead of cheering from the stands. Over time, you’re not just collecting wins—you’re training your beliefs to earn their keep.
To go deeper, here are 3 next steps: (1) Grab Karl Popper’s *Conjectures and Refutations* and, while you listen back to the episode’s segment on “good vs. bad hypotheses,” pause to translate one of your current beliefs into a Popper-style falsifiable statement (e.g., “If X is true, then I should observe Y in situation Z”). (2) Install a spaced-repetition tool like Anki and create a mini “falsifiability deck” with cards drawn from the episode’s examples (e.g., astrology, diet claims, productivity hacks), tagging each as “falsifiable” or “unfalsifiable” and testing yourself over the next week. (3) Open a free account on LessWrong and read their “A/B Testing and Falsifiability” or “Belief as Hypothesis to Test” posts, then set up a single, tiny real-world experiment today (like an A/B test on your email subject lines using Mailchimp or ConvertKit) that could actually prove your assumption wrong.
